KFBIO Thyroid Nodule Detection AI Solution Overview

Overview

The Challenge: Accelerating AI Diagnostics for Clinical Practice

As thyroid cancer cases continue to increase worldwide, the demand for faster and more accurate diagnosis is also growing. KFBIO developed an AI-powered thyroid nodule detection solution to support pathologists and improve diagnostic efficiency. However, before the solution could be widely adopted in clinical practice, its inference performance needed significant improvement.

To fit into routine pathology workflows, the system had to complete slide analysis in less than 180 seconds. At the same time, the engineering team needed to reduce hardware costs without sacrificing diagnostic accuracy.

To achieve these goals, KFBIO focused on three key technical challenges:

• Finding a cost-effective alternative to GPU-based inference
• Optimizing AI performance on CPU-based architectures
• Taking full advantage of the latest processor technologies for higher efficiency

The Solution: Optimizing AI Inference with Intel OpenVINO™

To overcome these challenges, KFBIO engineers integrated the Intel Distribution of OpenVINO™ toolkit into the AI inference workflow.

Instead of relying on expensive GPU-based processing, the team optimized the solution for CPU-based inference using 3rd Gen Intel® Xeon® Scalable processors. As a result, the AI model delivered much faster inference while maintaining high diagnostic accuracy.

OpenVINO™ also enabled the software to make better use of Intel processor resources. Consequently, the solution achieved substantial performance improvements without increasing hardware costs.

Performance Results

The optimized solution delivered measurable improvements across multiple performance indicators:

• 15.17× faster inference than non-optimized workloads on Intel® Xeon® Gold 6252 processors
• 21.38× higher overall performance on Intel® Xeon® 6330 processors
• Less than 180 seconds to complete analysis, meeting the requirements of routine clinical workflows

These improvements make the solution practical for real-world pathology applications, where both speed and accuracy are essential.

Improving Thyroid Nodule Detection with AI

Thyroid nodules have become an increasingly common healthcare concern worldwide. Although many nodules are benign, some can develop into thyroid cancer if they are not identified and treated early. Therefore, timely and accurate diagnosis plays a critical role in improving patient outcomes.

KFBIO’s AI solution helps pathologists detect thyroid nodules more efficiently while maintaining consistent diagnostic quality.

Advanced AI Technology

The solution combines advanced artificial intelligence with digital pathology technologies, including:

• Convolutional Neural Networks (CNNs) trained on thousands of medical images
• Seamless integration with digital pathology scanners and healthcare OEM devices
• A positive detection rate of more than 95% during validation testing

In addition, the AI model continuously delivers stable and reliable performance across different clinical scenarios, helping pathologists improve both efficiency and diagnostic confidence.

Benefits of Digital Pathology

Beyond AI-assisted detection, the solution also supports modern digital pathology workflows.

For example, it enables pathologists to review cases remotely and collaborate with specialists regardless of location. As a result, healthcare providers can respond more quickly to complex cases while improving diagnostic consistency.

The digital workflow also offers several additional benefits:

• Supports remote diagnosis and multidisciplinary collaboration
• Helps address the global shortage of experienced pathologists
• Expands access to high-quality pathology services in rural and underserved regions

Together, these advantages help improve both healthcare efficiency and patient access to expert diagnostic services.

Technical Implementation

To maximize inference performance, KFBIO completed a series of optimization steps.

First, the engineering team transitioned the AI inference architecture from GPU-dependent computing to CPU-optimized processing.

Next, they integrated the Intel Distribution of OpenVINO™ toolkit. This toolkit includes pre-optimized kernels and performance libraries that accelerate AI inference on Intel hardware.

Finally, the solution was deployed on 3rd Gen Intel® Xeon® Scalable processors. These processors provide enhanced AI acceleration capabilities and deliver higher overall computing efficiency.

By combining these three optimization strategies, KFBIO significantly improved inference speed while maintaining reliable diagnostic accuracy. As a result, the solution became well suited for routine clinical deployment in digital pathology laboratories.

Industry Impact and Future Applications

The successful deployment of KFBIO’s AI solution demonstrates how optimized computing technologies can accelerate digital pathology.

“Pathologists play a critical role in life-saving medical decisions, but global shortages limit access to expert diagnosis,” said Vito Wang, Director of Global Business Development at KFBIO.

“Our AI solution, accelerated by Intel technology, helps bridge this gap by enabling rapid and accurate thyroid nodule detection. As a result, it can be deployed even in resource-limited healthcare environments.”

More importantly, this collaboration highlights how AI and optimized computing architectures can improve both clinical efficiency and healthcare accessibility.

The solution delivers several key benefits:

• Reduces diagnostic costs through more efficient AI inference
• Extends expert pathology services to underserved regions
• Provides clinicians with fast, reliable, and consistent diagnostic support
• Improves workflow efficiency without increasing hardware investment

Furthermore, the project demonstrates that CPU-optimized AI inference can be a practical alternative to traditional GPU-based deployments for many pathology applications.

Conclusion

KFBIO’s collaboration with Intel has resulted in a scalable and highly efficient AI solution for thyroid nodule detection.

By integrating the Intel Distribution of OpenVINO™ toolkit with 3rd Gen Intel® Xeon® Scalable processors, KFBIO achieved up to 21.38× faster inference while maintaining high diagnostic accuracy.

As a result, the solution now meets the performance requirements of routine clinical workflows and supports wider adoption of AI-assisted pathology diagnosis.

In addition, this project demonstrates how software optimization and modern processor technologies can significantly improve AI performance while reducing deployment costs.

Looking ahead, the same optimization strategy can be applied to other digital pathology and medical AI applications. This approach provides a scalable framework for accelerating AI adoption across healthcare systems.

Configuration Details: Testing performed on Intel Xeon Gold 6330 processors. Performance results based on internal KFBIO benchmarking. Actual results may vary based on system configuration and workload characteristics.

Contact Information:

• Learn more about KFBIO: en.kfbio.cn
• Intel AI Builders Program: builders.intel.com/ai
• OpenVINO Toolkit Resources: software.intel.com/openvino-toolkit